The rainfall data of 20 rain gauge stations are used for analysis on the spatiotemporal characteristics of rainstorm-induced hazards in Beijing. A local model used to calculate rainstorm hazards index values (RHIVs) has been developed to reflect the degree of rainstorm-induced hazards. The Mann–Kendall test on the RHIVs series recognizes 1984 as the changepoint in the series, which is shortly after the beginning of a rapid urban expansion period in 1981. The RHIV trend analysis reveals that the trends indices of all stations are negative before 1984 but mostly positive after 1984. Although the climate in north China shows drought conditions, and the annual rainfalls have decreased in recent years, no relationship is implied to a reduction in rainstorm-induced hazards. By using the lognormal distribution model, the probability analysis on the RHIVs of 20 stations indicates that very extreme precipitation occurred in increasing frequency after 1984. Moreover, the estimated spatial distributions of 100-, 150-, and 200-mm daily rainfall exceedance probabilities (EPs) indicate that these EPs have increased mainly in the urban areas and northward, which are the downwind of the summer monsoon, whereas the EPs to the south of the urban areas have decreased since 1984. Such spatiotemporal characteristics of the RHIVs can be attributed to modification of precipitation by the changed land use and land cover in urban areas. Moreover, the urban-induced rainfall downwind of the urban areas enhanced the rain intensity and rain rate, which led to an increase in RHIVs and contributed to the frequent occurrences of flash floods in Beijing metropolis.
In recent years, heavy rains have fallen frequently in north China, which has led to severe flooding in many cities and has resulted in significant numbers of casualties and damage (Zhou et al. 2013). These events, which have raised a considerable amount of attention by the media worldwide, are thought to be the impacts of increased climate and weather extremes on the background of climate changes (Allen and Ingram 2002; Solomon et al. 2007; Bouwer 2011) combined with the vulnerability of densely populated areas and aggregated economic entities exposed in these flooded urban areas (Müller et al. 2011; Ashley et al. 2014). It has been reported that changes in land cover and land use as a result of urbanization can significantly affect the climate and environment in a metropolis (Changnon et al. 1991; Dixon and Mote 2003; Shepherd and Burian 2003; Shepherd 2005; Kaufmann et al. 2007; Ashley et al. 2012). In particular, increases in water vapor flux, land surface roughness, and aerosols are favorable for the enhancement of heavy rain (Shepherd et al. 2002; Pielke et al. 2007; Zhang et al. 2009; Rozoff et al. 2003), indicating that flash-flood hazards in urban areas can be consequently changed by urbanization.
The highly urbanized districts of Beijing, Tianjin, and Hebei in north China are typical areas undergoing the climate and synoptic changes induced by urbanization (Guo et al. 2006; Zhang et al. 2009). However, further investigation is needed to determine whether these changes have actually caused increases in rainstorm hazards and whether they can be used to explain the frequent occurrence of flash-floods in the urban areas of north China.
According to fourth report of the Intergovernmental Panel on Climate Change, assessment of the responses of hydroclimatic extremes to anthropogenic change is a main future uncertainty (Trenberth et al. 2007). Many previous studies reported that urban effects lead to increased precipitation during the summer months, and the increased precipitation is typically observed downwind of urbanized regions (Loose and Bornstein 1977; Braham et al. 1981; Changnon et al. 1991; Hand and Shepherd 2009). For example, Shepherd et al. (2002) employed precipitation radar data from the Tropical Rainfall Measuring Mission (TRMM satellite) to identify warm-season rainfall patterns around Atlanta, Georgia, and revealed an average increase in rainfall rates 30–60 km downwind of the city. By using a numerical model, Hjelmfelt (1982) simulated the urban heat island (UHI) in St. Louis, Missouri, and found positive vertical velocities downwind of the city. These results suggest that UHI intensity, increased urban surface roughness, and urban boundary layer instability enhance convection.
Although many studies have recognized the enhancement of convection downwind of cities (Shepherd 2005; Mote et al. 2007; Stallins and Rose 2008; Rose et al. 2008), urban effects on precipitation remain controversial, particularly in north China. Kaufmann et al. (2007) reported that urbanization weakened local precipitation in the Pearl River delta of China because of changes in surface hydrology that extend beyond the UHI effect and those in energy-related aerosol emissions. Guo et al. (2006) determined that in Beijing, the total accumulated precipitation in the entire domain decreased, particularly in the urbanized region, whereas a high precipitation intensity core remained downwind of the city. Yu et al. (2010) reported that the summer rainfall amount and frequency have significantly decreased in north China, whereas the rainfall intensity has increased. However, a decrease in rainfall amount and frequency does not necessarily correlate to a reduction in the risk of flash floods caused by heavy rain. Actually, except for extreme precipitation, local rainfall of short duration but high intensity can still result in flash flooding in urban areas because of enhanced rain intensity and rain rate as well as extended impervious land surfaces (Santosh et al. 2009; Smith et al. 1994). Obviously, changes in rainstorm-induced hazards in urban areas can be plausibly related to the effects of urban-induced rainfall and its modification of heavy rain. However, it is necessary to determine whether urban-induced rainfall has resulted in substantial enhancement of the rainstorm hazards, and the spatiotemporal changes in storm hazard in urban areas can be consequently recognizable.
In the research, storm frequency of a certain intensity is derived from the meteorological data, and its spatial distribution is plotted in a geographic information systems to determine the spatial variation and changes in storm hazards that are modified by urban-enhanced rainfall. Furthermore, the temporal changes in these hazards, which are influenced by the combination of climate change and urban-induced rainfall, are also analyzed by using a lognormal distribution model. An improved understanding of the spatiotemporal characteristics of storm hazards and their trends is critical in the recognition of thunderstorm risk in cities and is essential for urban safety and risk reduction.
The daily rainfall datasets recorded by 20 rain gauge stations in Beijing for the period from 1950 to 2012 have been used for the rainstorm hazard analysis. These data are provided by the Beijing Meteorological Bureau of the China Meteorological Administration. Most of the 20 observation stations within Beijing were established in the late 1960s and formed the fixed network by the end of that decade (Fig. 1). Some observations, such as Guanxiangtai, have consistent meteorological data over longer period before the 1980s. The datasets from the beginning of these stations observations, though their large sample sizes, are used in this paper.
To examine the relationship of rainstorm hazards and precipitation, data of historical damage were used for verifying flood losses caused by heavy rains with varied amounts of rainfall. Such analysis enabled the determination of storm hazard degree according to damage amount. The dataset used includes loss data related to floods occurring between 1981 and 2005 recorded by the Dictionaries of Meteorological Disasters in China (Beijing volume; Xie et al. 2005), and it is supplemented by data recorded in the Beijing Disaster Reduction Yearly Books from 2006 to 2010 (Beijing Disaster Reduction Association 2014). This information contains the date, time, and locations of these floods in addition to the numbers of fatalities and injuries and the amount of financial damage reported, which are the two basic factors in our risk assessment.
These historical loss data focused on the heavy floods and ignored some floods that had not caused extraordinary damage, so that it slightly underreported damage in total (Jiang et al. 2005). However, our work was to determine the correlation of precipitation to the damage index values. The negligence of these trivial floods could not affect our analysis fundamentally, because records of other similar floods in the reports could be used alternatively. The disadvantage of the data is that its annual economic losses are incomparable because of inflation (Jiang et al. 2005; Li et al. 2012). Thus, these loss data have been recalculated to that of the 2010 price level by using the 1989–2010 price indexes published in the Chinese Statistics Yearbooks (http://www.stats.gov.cn/tjsj/).
a. Rainstorm hazards index values (RHIVs) estimation
According to the Dictionaries of Meteorological Disasters, most historical floods were usually induced by rainstorms in which the daily rainfall amount exceeded 100 mm. However, under the conditions of vegetation reduction and impervious land surface extension, surface runoff can increase rapidly to sharply reduce the lagged time for peak streamflow. Such accelerated flow concentration obviously results in severe waterlogging or flash flooding in urban areas (Sheeder et al. 2002; Tang et al. 2005; Yang et al. 2009). Thus, the flash-flood threshold in terms of daily rainfall amount in urban areas should be adjusted to nearly 50 mm because of the extension of impervious land surface caused by urbanization (Arnold and Gibbons 1996; Gregory et al. 2006).
Figure 2 shows a correlation of precipitation to its potential loss plotted with the damage data of historical floods (Meyer et al. 2009); the damage index values were calculated by using a gray relational model (Wei 2010), estimating the weights related to the degree of gray relation among the multiple losses data of historical floods (Hu et al. 2011). The curve can be fitted approximately by the exponential function; thus, the hazard index value is defined as
where is the potential hazard threshold in terms of daily rainfall amount, which can cause flash flooding in urban areas and is usually set to 50 mm, and P is the daily rainfall and is divided by, given that a daily rainfall of more than 50 mm would lead to flash flooding in an urban area. The formula indicates that the RHIVs are in exponential augmentation by the nonlinear rather than linear scale. Theoretically, the rainfall amount cannot directly reflect the storm hazard degree; thus, the RHIVs can be used as a substitution.
The measurement of RHIVs in the yearly time series is considered as the sum of the hazard index values for each year on days in which the rainfall exceeded a specific threshold. Different from the definition of extreme event threshold in terms of a fixed percentile (Krishnamurthy et al. 2009; Beguería 2005), the rainstorm hazard threshold is determined by the minimal daily precipitation that can induce flash flooding in urban areas. The yearly time series of RHIVs is computed as
where t is the year, j is the rain gauge station, is the rain on day i in year t at gauge station j, and is the hazard threshold in terms of daily rainfall amount; κ is an indicator function that takes the value 1 if the argument is true and 0 otherwise.
Using rainfall data to estimate rainstorm hazards is critical in trend analysis and risk recognition. To represent the rainstorm hazard index, it is desirable to convert rainfall amounts to RHIVs in normalization (Beguería and Vicente-Serrano 2006) because this method measures the rainfall intensity and frequency according to flood probability and expected losses under certain precipitation conditions rather than that according to rainfall amount. Moreover, the hazard indices of all rainfalls over the threshold value in one year were summed to present the RHIVs in a yearly series, which actually considers the intensity and frequency together in the trend analysis. This method retains all heavy rain events in estimating RHIVs and avoids the omission of rain events that can potentially cause flash flooding even with relatively unremarkable rainfall amounts.
b. Exceedance probability and return period estimator
The probability of extreme rainfall can be estimated by using a lognormal distribution model. The lognormal probability density function is described as
where μ is the scale parameter, and σ is the shape parameter (Evans et al. 1993), which can be calculated by
Then, its cumulative distribution function is
and the exceedance probability (EP) of the threshold value x0 can be expressed as
According to Eq. (1), the threshold value x0 is
Then, its corresponding rainfall P0 is
Therefore, the return period of the threshold value of rainfall P0 can be calculated as
c. RHIV series trend analysis
The characteristics of heavy rains are determined by the climate and are affected by factors of climate change including anthropologic changes such as weather modified by urbanization (Changnon and Demissie 1996; Kaufmann et al. 2007). Thus, it is necessary to verify storm hazard inconsistency in the yearly series of certain climate. Determination of changepoints in the RHIV series enables easy recognition of RHIV trends and is useful in risk prediction. Therefore, the RHIVs of each rain gauge station were calculated in this study by using rainfall data and were validated against the changepoints in the RHIV series. These changepoints can be determined by applying the nonparametric Mann–Kendall (MK) test (Yang et al. 2009), and the changes in rainstorm hazards in the RHIV series can be highlighted to be used in trend analysis.
The MK test is a rank-based test with no assumptions as to the underlying probability distribution of data (Helsel and Hirsch 1992). In this test, given the time series x composed of n samples, a rank sequence can be initialized as
where the rank sequence sk is the count of the ith value xi greater than the jth value xj.
Given the independent stochastic process of the time series, the statistic UF can be defined as
where UF1 = 0, E(sk) and var(sk) are the mean and variance of the accumulative value sk, respectively.
Following the diverse sequence of the time series x, which is xn, xn−1, …, and x1, repeat the procedure of Eqs. (11)–(13), and let UBk = −UFk (k = n, n −1, …, 1), UB1 = 0. If UFi > 0 and UBi < 0, the values in the time series present downward trends. Otherwise, if UFi < 0 and UBi > 0, the values present upward trends. However, if the lines of UF and UB cross at the kth value, and the cross point is between the critical lines, the kth time of the cross point is the changepoint in the series.
4. Analysis of RHIVs
a. Annual rainfall histogram of rain gauge stations
Figure 3 shows the annual rainfall histograms of Guanxiangtai, Haidian, and Miyun, which are typical rain gauge stations located south, center, and north of the Beijing metropolis, respectively. These stations were established in 1951, 1975, and 1971, respectively, and the Guanxiangtai station has the longest term of rainfall data. The combined display of these rainfall data histograms indicates that the overall annual rainfall in Beijing presents downward trends, implying that the climate of Beijing tends to be drought. At Guanxiangtai, for example, the mean annual rainfall was 576 mm in 1951–2012, 594 mm in 1951–84, and 563 mm in 1984–2012. The decrease in annual rainfall from the beginning of rapid urbanization is in agreement with that reported by Guo et al. (2006), Zhang et al. (2009), and Yu et al. (2010), all of whom determined that the precipitation in Beijing has decreased in the past 20 yr.
b. Trends analysis
1) Determination of the changepoint by using MK test
Daily rainfall amounts exceeding 50 mm or the peak daily rainfall amount in one year if no daily rainfall exceeded 50 mm were converted to RHIVs by using Eq. (2) and were subsequently initialized as RHIVs series for analysis.
The MK test is used to detect the changepoint in the RHIV series of Guanxiangtai because that station has the longest-term observation data in the Beijing district, which are more suitable for trend analysis. Figure 4 shows the lines of UB and UF determined by the MK test. The cross points of UB and UF were in close proximity around the year of 1984 in an oscillating manner; therefore, the changepoint should have been in 1984. Prior to 1984, the UBs were negative, indicating that the RHIVs tended to show decreases at that time. After 1984, the UBs fluctuated between negative and positive but were mostly positive and showed an abruptly increasing tendency; therefore, the RHIVs presented upward trends in general after 1984.
It is noticeable that 1984 would have been the changepoint in the RHIV series. The “Beijing Statistical Annals” (http://www.bjstats.gov.cn) show that Beijing entered a rapid urban expansion period after 1981 (Zhang et al. 2009). In the 1980s, the urban areas were mostly enclosed by the second ring road (Wang et al. 2008); however, these urban areas have since expanded to the fifth ring road (Fig. 1). The beginning of the RHIV increase corresponds to the beginning of rapid urban expansion shortly before 1984. Thus, it is unclear whether this RHIV trend correlated to the urbanization.
Results of Haar wavelet analysis on the RHIVs of Guanxiangtai also indicated an apparent changepoint in the RHIV series in approximately 1981–84 (Fig. 5), which is similar to that determined by the MK test; 2000 and 2008 were determined as secondary changepoints in both tests.
2) Linear trend analysis
In comparison of the RHIVs trends before and after 1984, the linear trend indices of the 20-station RHIV series should be estimated. These trends can be described as
where yi is the ith year in the RHIV series, ti is the dependent variable, a is the intercept, and b is the trends index. The linear trends were fitted by using the least squares method. Figure 6 shows the trend index histogram of the 20 stations in the RHIV series before and after 1984. All trend indices b of these stations were negative before 1984. Conversely, those of most stations (almost 13) were positive after 1984, and the other negative stations, such as Guanxiangtai and Fengtai, are located mostly south of urban areas. These results verify that 1984 is the changepoint in the RHIV series of Beijing. After 1984, the RHIVs of the stations mostly downwind of urban areas presented upward trends, whereas those of some stations to the south presented downward trends.
Generally, the RHIVs appear to have not peaked before 2012. Moreover, the RHIVs trends are positive, and the RHIVs will continue to increase; therefore, rainstorm hazards in the future will become more serious than ever before.
The spatial distribution of the RHIV trend indices of the 20 stations before and after 1984 (Fig. 7) were also plotted by using the interpolation of inverse distance weighting method (IDWM). This method clarifies the linear trends of the RHIVs, in which the positive trend index presents upward trends. A greater value relates to a more increasing degree of RHIV; otherwise, it presents downward trends. The spatial distributions of RHIV trends before 1984 show that the trend indices were all negative and that the negative center was slightly to the north of the urban area (Fig. 7a). These results indicate that the RHIVs in Beijing presented downward trends before 1984. For distributions of the trend indices after 1984 (Fig. 7b), those over and downwind of the urban area were positive, and the peak center was slightly to the north of the urban area, indicating the RHIVs of these areas presented upward trends. On the contrary, the trend indices to the south and the north far from the urban area were negative.
c. Probability of extreme precipitation
The RHIV series of 20 stations derived by daily rainfall datasets also served as samples for probability analysis by using the lognormal distribution model. The 100-, 150-, and 200-mm (daily rainfall) cumulative probability EP and return period were estimated by using Eqs. (2)–(10) with rainfall data of two observation periods. The first period included all available data that have been observed since these stations were established, and the second included those observed after 1984, corresponding to the changepoint in the RHIV series. All of the estimate parameters for the lognormal distribution of all stations in the RHIV series passed the 95% confidence level.
Figure 8a shows that the 100-mm daily rainfall EPs of most stations (85%) estimated by using all available data were greater than those using only data after 1984, which implies that the 100-mm daily rainfall EPs of the stations after 1984 were lower than those during all observation periods.
At most stations, the EPs of heavier rainfall after 1984 were greater relative to the EPs during all observation periods (Figs. 8b,c); that is, the 100-mm daily rainfall EPs of three stations after 1984 were greater than those during all observation periods, whereas the 150-mm daily rainfall EPs of 6 stations and the 200-mm daily rainfall EPs of 11 stations (55% of the total stations) were greater. Therefore, in Beijing, the 100–150-mm daily rainfall EPs before 1984 were relatively greater than those after 1984, whereas the EPs of the 200-mm (daily rainfall) extreme precipitation after 1984 have become greater than those before 1984.
Research has reported the effects of aerosols, which can suppress precipitation from shallow clouds in polluted areas (Ramanathan et al. 2001; Shepherd 2005; Rosenfeld et al. 2008). Beijing has become polluted since the beginning of the accelerated urbanization in 1984, which suggests the decrease of 100–150-mm daily rainfall EPs could be caused by the effects of aerosols on precipitation; additional research outside the scope of this manuscript is required to confirm this hypothesis.
To reveal the temporal changes in RHIVs after and before 1984, I calculated the difference values (DVs) among 100-, 150-, and 200-mm daily rainfall EPs that were estimated by using only data observed after 1984 EPAfter1984 and that by using all available data EPAll:
Then, the spatial distributions of 100-, 150-, and 200-mm (daily rainfall) DVs were plotted by using IDWM interpolation for comparison (Fig. 9). The 100-mm DVs to the north of the urban areas were relatively greater than those in any other area and were partially in positive values; the 100-mm DVs positive peak center was located in the neighborhood of Miyun (Fig. 9a). The 150-mm DVs positive regions were spread to the north of the urban area (Fig. 9b). Similarly, the 200-mm DVs positive regions were mostly to the north of the urban area (Fig. 9c), although the peak center was in the Jumahe River valley, where the orographic uplift can enhance convection. In contrast, the 100-, 150- and 200-mm DV low centers in negative values were mostly located to the south of urban areas (Fig. 9).
The area upwind of the summer monsoon in Beijing is to the south of the urban area, and that downwind is to the north (Sun and Shu 2007). The rainstorm probability of “high in north and low in south” is possibly associated with the urban-induced rainfall, in which the precipitation is modified by urban land use and land cover, resulting in precipitation enhancement downwind of urban areas (Pielke et al. 2007). An area of high lightning density to the north of urban areas in Beijing had also been detected in the previous work (Hu et al. 2014); it is considered to be effect of urban modification on thunderstorms (Stallins 2004; Stallins and Rose 2008; Ashley et al. 2012; Rose et al. 2008). The rainstorm EPs to the north of the urban areas were greater than those to the south. From this perspective, the extreme precipitation likely had been modified by the changes of land use and land cover in urbanization.
In Beijing, the mean annual rainfall before 1984 is apparently greater than that after 1984, and the climate tends to be drought, in agreement with the so-called south flooding and north drought in east China (Yu et al. 2010). In recent years, the cooled top layer of the troposphere in summer in north China has weakened the summer monsoon, whereas the lower atmosphere was warmed remarkably because of climate changes owing to global warming (Easterling et al. 2000) and the UHI effect (Jiang and Liu 2006; Sun and Shu 2007). Thus, the static stability is weakened, and the convection can be easily stimulated in the afternoon or evening. However, the weakened summer monsoon is unable to bring sufficient water vapor to north China; thus, it cannot reach the vapor conditions of persistent precipitation, and the rainfall subsequently decreases. Although the rainfall totals in north China have decreased, the rain intensity or rain rate will increase and frequently cause flooding in urban areas, which brings about a surge of rainstorm hazards.
The MK test demonstrated that 1984 was the changepoint in the RHIV series; similar results were obtained in the Haar wavelet analysis. Although this changepoint occurred shortly after the rapid urbanization in Beijing, this evidence is insufficient to conclude that the changepoint in 1984 is related to the urbanization. Such a perspective should be supported by stronger evidence.
Considering 1984 as the changepoint in the RHIV series, the linear trend indices estimated by using the least squares method also prove that the RHIVs of all stations presented downward trends before 1984. After 1984, however, the RHIVs of most stations presented upward trends, and those showing downward trends are located to the south of urban areas, upwind of the summer monsoon in Beijing.
The spatial distributions of 100-, 150-, and 200-mm DVs indicate that after 1984 the increased extreme precipitation EPs areas, in which the daily rainfall amount exceeded 150 mm, were mainly in the urban areas and to the north, downwind of the summer monsoon. However, the extreme precipitation EPs to the south of the urban areas decreased and presented downward trends after 1984. The analysis also revealed that the 100-mm daily rainfall EPs after 1984 were usually lower than those before 1984; nevertheless, the 150-mm daily rainfall EPs increased slightly until the extreme 200-mm daily rainfall EPs after 1984 became relatively greater than those before 1984. These results demonstrate that the risk of extreme precipitation increased correspondingly after 1984.
These results can be attributed to precipitation that would have been modified by the urban environments. Therefore, the urban-induced precipitation would have enhanced the rainfall amount and intensity and consequently resulted in the increase of RHIVs in Beijing. Possible mechanisms for the urban environment effect on precipitation or convection have been well summarized by Shepherd (2005), for example, “enhanced convergence, destabilization, enhanced aerosols in the urban environment, and bifurcating or diverting of precipitating systems by the urban canopy.” Moreover, Jiang and Liu (2006) and Sun and Shu (2007) argued that in Beijing, the existence of warm centers in urban areas results in relatively lower pressure and drives the convergence of ascending cooler air in the suburbs, which is favorable for producing local heavy rain. Such results reflect the increase in RHIVs.
A decrease in annual rainfall does not necessarily indicate a decrease in RHIVs. The findings indicate that precipitation can be modified by the land use and land cover in urban areas and that the urban-induced rainfall downwind of urban areas enhances the rain intensity and rain rate, which results in an increase in RHIVs. Such results would contribute to the frequency of flash-flood occurrences in the Beijing metropolis. Additionally, Yu et al. (2010) concluded that the rainfall intensity has increased in north China, although the rainfall amount and frequency have significantly decreased. These RHIV trends also have been demonstrated by the frequent occurrences of extreme precipitation in recent years.
On the other hand, urbanization leads to aggregation of populations and economic entities in cities, where they are concentrated, exposed, and vulnerable to disasters (Ashley et al. 2014; Hu 2014), especially in the urbanized areas on floodplains or wetlands (Shi et al. 2005). In Beijing, the population increased from 9.65 million in 1984 to 20.693 million in 2012 while the urban areas expanded from 360 to 1268 km2 (http://data.stats.gov.cn/workspace/index?m=hgnd), with an annual increment of 32.429 km2. The extension of these impervious surfaces makes the communities more susceptible to flash floods in urban areas (Ogden et al. 2000). Thus, in addition to the likely increase in hazards induced by urban-enhanced precipitation, future research of severe weather risk management must recognize the risk of flash flooding in urban areas to reduce risk and thus avoid disasters.
All of the available rainfall data of 20 rain gauge stations are used for analysis on the RHIV spatiotemporal characteristics in Beijing. The MK test on the RHIV series is used to determine that the changepoint in the series was 1984, shortly after the beginning of the rapid urban expansion since 1981 in Beijing. The RHIVs of all stations present downward trends before 1984, and the RHIVs of most stations present trends that move upward in an oscillating manner after that time. The upward-trend stations are in close proximity to the urban areas or to the north, downwind of the summer monsoon, whereas the downward-trend stations are to the south of the urban areas, upwind of the summer monsoon.
Probability analysis on the 20 stations performed by using the lognormal distribution model reveals that the 100-mm daily rainfall EPs of 20 stations after 1984 could be lower than those before 1984. In addition, the 150-mm daily rainfall EPs have increased slightly, although most are lower, and the 200-mm daily rainfall EPs after 1984 are greater than those before that time, which indicates that very extreme precipitation has increased in frequency after 1984. The spatial distributions of 100-, 150-, and 200-mm DVs reflect that the increased areas of 100-, 150-, and 200-mm daily rainfall EPs are mainly in the urban areas and to the north, downwind of the summer monsoon.
Although annual rainfall has significantly decreased in recent years, urban-induced rainfall has enhanced the precipitation downwind of the urban areas. The modification of precipitation has also influenced extreme rainfalls and has resulted in changes in rainstorm hazards in Beijing. Consequently, this modification has led to upward trends of RHIVs in the urban areas and to the north, whereas the downward trends of RHIVs are to the south of the urban areas.
Thus, it is illogical to assume that the annual rainfall in north China has decreased and the climate tends to be drought, so the rainstorm risk is reduced accordingly. Actually, the precipitation downwind of the summer monsoon could have been modified by changes in land use and land cover in urbanization, which results in the enhancement of rain intensity and rain rate in these areas. Moreover, the increased areal coverage of impervious surfaces in urban areas can alter natural hydrologic responses. The RHIVs in the Beijing metropolis can be subsequently enhanced by the precipitation modification and hydrologic effects of these extended impervious surfaces, along with the amplified exposure and vulnerability to disasters. Therefore, it is imperative to reinforce monitoring and warning of severe weather in these regions to promote safety and reduce risk.
This study has been supported by the National Natural Science Foundation of China (Project 41175099) and the Beijing Natural Science Foundation of China (Project 8142019).